Fix v4 token-count arithmetic: v4 data build is 257.9M tok (fw 150.6M + dclm 107.3M), cumulative ~308M, ~30 tok/param, 8.4x shortfall — not the previously stated ~393M/~444M/~43.1. Verified against models/ldt-10m-v4/train.log data-build lines.
#12
by Compactbot - opened
No description provided.
Verified token-count correction. v4 data build (from train.log): fw 150,628,269 + dclm 107,295,779 = 257,924,048 tok; cumulative 308M; ~30 tok/param; 8.4x shortfall vs the 2.6B request. The previously stated ~393M/444M/~43.1 figures were an arithmetic error.
Compactbot changed pull request status to merged